Spark中DataFrame与DataSet的使用

概述：本文主要介绍了DataFrame、DataSet常见API的使用

1、DataFrame的使用

object DataFrameApp {
  def main(args: Array[String]): Unit = {
    val spark = SparkSession.builder().master("local[2]").getOrCreate()
​
    //将json转为DataFrame
    val peopleDF = spark.read.format("json").load("file:///F:/people.json") 
   //hdfs://hadoop:8020/input/wc/people.json
    peopleDF.printSchema()
    //默认输出前20条记录
    peopleDF.show()
​
    //查询某列数据
    peopleDF.select("name", "age").show()
    //查询列数据并进行计算
    peopleDF.select(peopleDF.col("name"), (peopleDF.col("age") + 9).as("newAge"))
​
    //条件查询
    peopleDF.filter(peopleDF.col("age") > 19).show()
    peopleDF.where(peopleDF.col("age") > 19).show()
​
    //分组与聚合操作 select age,count(1) from table group by age
    peopleDF.groupBy("age").count().show()
    spark.stop()
  }
}

2、RDD与DataFrame的互操作

（1）反射方式

object DataFrameRDDApp {
  def main(args: Array[String]): Unit = {
    val spark = SparkSession.builder().appName("DataFrameRDDApp").master("local[2]").getOrCreate()
    //RDD转DataFrame
    val rdd = spark.sparkContext.textFile("file:///F:/people.txt")
    //toDF需要导入隐式转换
    import spark.implicits._
    val peopleDF = rdd.map(_.split(",")).map(line => people(line(0).toInt, line(1).toString, line(2).toInt)).toDF()
    peopleDF.printSchema()
    peopleDF.show()
    peopleDF.filter(peopleDF.col("age")<25).show()
    //创建临时视图 使用sql操作数据库
    peopleDF.createOrReplaceTempView("peopleTmpView")
    spark.sql("select p.id ID,p.name USERNAME,p.age AGE from  peopleTmpView p WHERE name LIKE '%an%'").show()
    spark.stop()
  }
  case class people(id: Int, name: String, age: Int)
}
​

（2）编程方式

使用情形：

• When case classes cannot be defined ahead of time (for example, the structure of records is encoded in a string, or a text dataset will be parsed and fields will be projected differently for different users).

创建DataFrame步骤：

• Create an RDD of Rows from the original RDD;
• Create the schema represented by a StructType matching the structure of Rows in the RDD created in Step 1.
• Apply the schema to the RDD of Rows via createDataFrame method provided by SparkSession.

代码：

object DataFrameRDDApp {
  def main(args: Array[String]): Unit = {
    val spark = SparkSession.builder().appName("DataFrameRDDApp").master("local[2]").getOrCreate()
    //RDD转DataFrame
    val rdd = spark.sparkContext.textFile("file:///F:/people.txt")
    val peopleRDD = rdd.map(_.split(",")).map(line => Row(line(0).toInt, line(1).toString, line(2).toInt))
    val structType = StructType(Array(StructField("id", IntegerType, true), StructField("name", StringType, true), StructField("age", IntegerType, true)))
    val peopleDF = spark.createDataFrame(peopleRDD, structType)
    peopleDF.printSchema()
    peopleDF.show()
    peopleDF.filter(peopleDF.col("age") < 25).show()
    //创建临时视图 使用sql操作数据库
    peopleDF.createOrReplaceTempView("peopleTmpView")
    spark.sql("select p.id ID,p.name username,p.age AGE from  peopleTmpView p WHERE name LIKE '%li%'").show()
    spark.stop()
  }
}

模糊查询结果：

选型说明：如果字段名称、类型可知，优先选择反射方式

（3）DataFrame其他API

object DataFrameCase {
  def main(args: Array[String]): Unit = {
    val spark = SparkSession.builder().appName("DataFrameCase ").master("local[2]").getOrCreate()
    //RDD转DataFrame
    val rdd = spark.sparkContext.textFile("file:///F:/student.data")
    import spark.implicits._
    val studentDF = rdd.map(_.split("\\|")).map(line => student(line(0).toInt, line(1).toString, line(2).toString, line(3).toString)).toDF()
    studentDF.first()
    studentDF.printSchema()
    studentDF.show(30,false)
    studentDF.filter("name='' OR name='null'")
    //查找以M开头的学生
    studentDF.filter("SUBSTR(name,0,1)='E'").show()
    studentDF.sort(studentDF("name")).show()
    studentDF.sort("name","email").show()
    studentDF.sort(studentDF("name").desc).show()
    studentDF.select(studentDF("name").as("studentName")).show()
    //默认内连接 需要使用三个=
    val newStudentDF = rdd.map(_.split("\\|")).map(line => student(line(0).toInt, line(1).toString, line(2).toString, line(3).toString)).toDF()
    studentDF.join(newStudentDF,studentDF("id")=!=newStudentDF("id"),"right").show(500,false)
    studentDF.take(10).foreach(println)
    studentDF.first()
    studentDF.head(5)
    spark.stop()
  }
  case class student(id:Int,name:String,phone:String,email:String)
}

根据姓名排序结果：

3、DataSet的使用

（1）DataFrame转DataSet

object DataSetApp {
  def main(args: Array[String]): Unit = {
    val spark = SparkSession.builder().appName("DataSetApp").master("local[2]").getOrCreate()
    val path = "file:///F:/student.csv"
    //解析csv文件 外部数据源功能
    val df = spark.read.option("header", "true").option("inferSchema", true).csv(path)
    df.show()
    //隐式转换 df转ds
    import spark.implicits._
    val ds = df.as[student]
    ds.map(line=>line.email).show()
    spark.stop()
  }
  case class student(id: Int, studentName: String, phone: String, email: String)
}

DataSet的查询邮箱结果：

（2）DataSet静态类型（Static-typing）和运行时类型安全（runtime type-safety）

说明：

• spark.sql("seletc * from emp e join dept d on e.deptno=d.deptno").show()编译可以通过，执行sql时报错
• studentDF.seletc(studentDF("name").as("studentName")).show()中编译将不能通过
• ds.map(line=>line.emails).show()中编译将不能通过，Dataset不管是哪种状态都是安全的，即静态类型和运行时类型安全，因为Dataset两种状态都能把错误抛出来。